NO apologies will ever be given for that stance. I expect the best from people working for me and you should also.
http://www.cell.com/trends/neurosciences/fulltext/S0166-2236(17)30237-0
,
Published Online: December 21, 2017
Open Access
Hide Pane
l
Highlights
Synaptic remodeling is driven by both activity-dependent and spontaneous processes.(If your doctor doesn't figure out how this can help you recover you don't have a doctor, maybe a wooden post?)
The magnitude of spontaneous synaptic remodeling is comparable with that of activity-dependent remodeling.
Spontaneous synaptic remodeling processes can give rise to a full repertoire of synaptic sizes even in the complete absence of activity.
While spontaneous remodeling processes drive continuous fluctuations in the properties of individual synapses, stable and skewed distributions of the same properties emerge as population invariants.
Scaling of synaptic size distributions appears as a population-level phenomenon when spontaneous remodeling processes are modulated by perturbations of spontaneous network activity.
The magnitude of spontaneous synaptic remodeling is comparable with that of activity-dependent remodeling.
Spontaneous synaptic remodeling processes can give rise to a full repertoire of synaptic sizes even in the complete absence of activity.
While spontaneous remodeling processes drive continuous fluctuations in the properties of individual synapses, stable and skewed distributions of the same properties emerge as population invariants.
Scaling of synaptic size distributions appears as a population-level phenomenon when spontaneous remodeling processes are modulated by perturbations of spontaneous network activity.
Synaptic
plasticity – the directed modulation of synaptic connections by
specific activity histories or physiological signals – is believed to be
a major mechanism for the modification of neuronal network function.
This belief, however, has a ‘flip side’: the supposition that synapses
do not change spontaneously in manners unrelated to such signals.
Contrary to this supposition, recent studies reveal that synapses do
change spontaneously, and to a fairly large extent. Here we review
experimental results on spontaneous synaptic remodeling, its relative
contributions to total synaptic remodeling, its statistical
characteristics, and its physiological importance. We also address
challenges it poses and avenues it opens for future experimental and
theoretical research.
Synaptic Plasticity and Synaptic Tenacity
At
a conceptual level, the central nervous system (CNS) comprises a vast
network of excitable cells connected by specialized junctions known as
synapses. Long before excitability and synaptic transmission were fully
understood, scholars such as William James [1
suggested that changes to connections (‘tracts of conduction’), driven
by sequential or concomitant activation of ‘elementary brain-processes’,
might explain the formation of new associations and the learning of new
tasks. In fact, James credits this idea to yet earlier scholars (e.g.,
Descartes, Locke) and notes that they ‘hit upon this explanation, which
modern science has not yet succeeded in improving’ [1.
Years later, Donald Hebb rephrased this idea, using it as a cornerstone
for constructs such as cell assemblies, thereafter used to explain
network function and ultimately behavior in physiological terms [2.
Inspired by these theories, and armed with increasingly powerful
methodologies, numerous investigators have convincingly demonstrated
that activity histories (see Glossary)
can influence and change such tracts of conduction, now largely
identified as chemical synapses. The resulting changes, generally
referred to as synaptic plasticity, are thus commonly viewed as evidence
for this venerable theory.
The long-held framework described above has an implicit yet often ignored corollary, which is the supposition that synapses, when not driven to change by particular activity patterns or histories, retain their properties indefinitely. This supposition would seem to be essential because without it, spontaneously occurring changes might drive spurious modifications in network function or undo physiologically relevant ones. Thus, to fully appreciate the importance and limitations of synaptic plasticity, it is essential to measure and understand the capacity of synapses to maintain their particular properties (e.g., probability of release, total receptor conductance, size, morphology, ultrastructure, composition) over behaviorally relevant timescales. We refer to this capacity as synaptic tenacity [3, 4, 5, 6].
The aim of this Opinion is to discuss challenges to the notion of synaptic tenacity that come from general biological considerations and experimental findings. Such findings collectively suggest that synaptic tenacity is inherently limited, since synapses do change spontaneously and to a fairly large extent. We survey these findings and discuss their implications in light of the supposition mentioned above. We then present principles emerging from recent literature on the manners by which spontaneous processes might govern dynamic and statistical properties of synaptic populations. We end by formulating several key questions concerning relationships between spontaneous remodeling, underlying biological processes, neuronal activity, and network function.
The long-held framework described above has an implicit yet often ignored corollary, which is the supposition that synapses, when not driven to change by particular activity patterns or histories, retain their properties indefinitely. This supposition would seem to be essential because without it, spontaneously occurring changes might drive spurious modifications in network function or undo physiologically relevant ones. Thus, to fully appreciate the importance and limitations of synaptic plasticity, it is essential to measure and understand the capacity of synapses to maintain their particular properties (e.g., probability of release, total receptor conductance, size, morphology, ultrastructure, composition) over behaviorally relevant timescales. We refer to this capacity as synaptic tenacity [3, 4, 5, 6].
The aim of this Opinion is to discuss challenges to the notion of synaptic tenacity that come from general biological considerations and experimental findings. Such findings collectively suggest that synaptic tenacity is inherently limited, since synapses do change spontaneously and to a fairly large extent. We survey these findings and discuss their implications in light of the supposition mentioned above. We then present principles emerging from recent literature on the manners by which spontaneous processes might govern dynamic and statistical properties of synaptic populations. We end by formulating several key questions concerning relationships between spontaneous remodeling, underlying biological processes, neuronal activity, and network function.
Synaptic Tenacity: Inherent Challenges
CNS
synapses are micrometer-size, intricate assemblies of diverse molecules
(e.g., receptors, ion channels, synaptic vesicle, scaffolding,
cytoskeletal, adhesion, and signaling molecules). Typical lifetimes of
synaptic molecules (i.e., days; see [7)
are orders of magnitude shorter than the lifetimes of many, if not most
CNS synapses (weeks and months, and in some species probably years [8, 9, 10]; see [11
for a recent review). Consequently, the maintenance of synaptic
connections involves continuous removal and degradation of molecules and
their substitution with freshly synthesized copies. Given that CNS
synapses are often remote from the neuron’s somatic biosynthetic
machinery (residing at distances of many centimeters or even meters, in
some organisms), long-term synaptic maintenance is a remarkable
biological feat [12, 13].
Importantly, however, given these logistic challenges and the low copy
numbers of many molecules at individual synapses, it is probably
unrealistic to expect that synapses maintain their particular contents
and, by extension, their functional properties with pinpoint precision
(see [14).
This expectation is further challenged by the fact that synapses are
not rigid structures but rather are dynamic assemblies of molecules (and
organelles) that continuously migrate into, out of, and between
neighboring assemblies through lateral diffusion, active trafficking,
endocytosis, and exocytosis [15, 16].
These molecular dynamics are often further accelerated by cellular
processes associated with synaptic transmission, such as membrane
recycling and cytoskeletal dynamics.
Conceivably, challenges to synaptic tenacity posed by continuous molecular turnover and the myriad molecular dynamics might be met somehow, resulting in roughly stable synaptic properties. More likely, however, molecular dynamics coupled with imprecise proteostasis will drive spontaneous changes in synaptic properties that have little to do with specific activity histories [17. As we show next, experimental evidence seems to favor the latter possibility.
Conceivably, challenges to synaptic tenacity posed by continuous molecular turnover and the myriad molecular dynamics might be met somehow, resulting in roughly stable synaptic properties. More likely, however, molecular dynamics coupled with imprecise proteostasis will drive spontaneous changes in synaptic properties that have little to do with specific activity histories [17. As we show next, experimental evidence seems to favor the latter possibility.
Spontaneous Synaptic Remodeling in Developing Networks
During
development the brain is strongly influenced by experience and
activity, yet it seems that activity is largely dispensable for synapse
formation per se. Although this conclusion is well established [18, 19] (see [20 for a review of earlier work), recent studies have highlighted observations that are particularly pertinent to this Opinion.
Studies in primary culture have shown that general properties of excitatory and inhibitory synapses are largely preserved even when networks develop in the complete absence of activity. For example, chronic suppression of neurotransmitter release or network activity in primary cultures of hippocampal neurons does not appear to affect synaptic spatial densities, colocalization of pre- and postsynaptic molecules, or the synaptic contents of important postsynaptic density (PSD) proteins (e.g., PSD-95/Dlg4, CaMKIIα, SynGAP, Gephyrin) and neurotransmitter receptor subunits (Gria2/3, GABAAα2) [18, 19]. Recently, these findings were confirmed and extended to additional preparations, including the intact brain. In one study [21 (see also [22), spine types, spine spatial densities, and synaptic currents were examined in hippocampal organotypic cultures prepared from mice in which presynaptic release was abolished (by eliminating the presynaptic proteins Munc13-1 and Munc13-2). These synaptic features were found to be similar to those observed in preparations from control animals, although postsynaptic currents were slightly reduced and less correlated with spine volume. In a second study [23, the near-complete suppression of presynaptic glutamate release in the mouse forebrain (by tetanus toxin expression) was found to have relatively modest effects on synapses and their properties, at least as judged by morphological measures. Although synapse numbers and spine densities were reduced in some brain regions (but not others), dendritic spines of all types formed at normal proportions.
These observations – that synapses of all sizes and types form at normal proportions – have profound implications: when synapses and dendritic spines are initially formed, they generally have small volumes and PSDs (e.g., [24, 25, 26, 27]). Their subsequent conversion into large, mushroom-shaped spines is often thought to be driven by activity-dependent potentiation (reviewed in [28, 29]; see also [30, 31]). However, as mentioned above, large synapses, including mushroom-shaped spines, develop in normal proportions even when activity is essentially nonexistent [18, 19, 21, 23]. It thus seems that activity-independent processes can generate the full repertoire of synaptic sizes [17, highlighting the potency of spontaneous synaptic remodeling.
Studies in primary culture have shown that general properties of excitatory and inhibitory synapses are largely preserved even when networks develop in the complete absence of activity. For example, chronic suppression of neurotransmitter release or network activity in primary cultures of hippocampal neurons does not appear to affect synaptic spatial densities, colocalization of pre- and postsynaptic molecules, or the synaptic contents of important postsynaptic density (PSD) proteins (e.g., PSD-95/Dlg4, CaMKIIα, SynGAP, Gephyrin) and neurotransmitter receptor subunits (Gria2/3, GABAAα2) [18, 19]. Recently, these findings were confirmed and extended to additional preparations, including the intact brain. In one study [21 (see also [22), spine types, spine spatial densities, and synaptic currents were examined in hippocampal organotypic cultures prepared from mice in which presynaptic release was abolished (by eliminating the presynaptic proteins Munc13-1 and Munc13-2). These synaptic features were found to be similar to those observed in preparations from control animals, although postsynaptic currents were slightly reduced and less correlated with spine volume. In a second study [23, the near-complete suppression of presynaptic glutamate release in the mouse forebrain (by tetanus toxin expression) was found to have relatively modest effects on synapses and their properties, at least as judged by morphological measures. Although synapse numbers and spine densities were reduced in some brain regions (but not others), dendritic spines of all types formed at normal proportions.
These observations – that synapses of all sizes and types form at normal proportions – have profound implications: when synapses and dendritic spines are initially formed, they generally have small volumes and PSDs (e.g., [24, 25, 26, 27]). Their subsequent conversion into large, mushroom-shaped spines is often thought to be driven by activity-dependent potentiation (reviewed in [28, 29]; see also [30, 31]). However, as mentioned above, large synapses, including mushroom-shaped spines, develop in normal proportions even when activity is essentially nonexistent [18, 19, 21, 23]. It thus seems that activity-independent processes can generate the full repertoire of synaptic sizes [17, highlighting the potency of spontaneous synaptic remodeling.
Spontaneous Synaptic Remodeling in Established Networks
Longitudinal observations of individual CNS synapses over many days in mature networks, both in vitro and in vivo, indicate that most synapses are persistent over these timescales, (e.g., [10, 32] but see [9; reviewed in [8), although significant synapse formation and elimination are also observed (reviewed in [33, 34, 35]). Closer examination reveals, however, that properties of individual synapses, such as spine volume (e.g., [17, 24, 36, 37, 38]), presynaptic bouton volume [39, 40], synaptic vesicle number [4, 5, 41], active zone (AZ) molecule content (Bassoon, Munc13-1) [4, 42, 43], and PSD protein content (PSD-95, Gria2, Gephyrin, GKAP, Shank, Zip45/Homer1c) [3, 5, 6, 27, 43, 44, 45, 46, 47, 48, 49] fluctuate considerably over these timescales (as illustrated for PSD-95 in Figure 1). Given the strong correlations between these measures and functional synaptic features such as synaptic vesicle release [42 and synaptic current amplitudes (reviewed in [8, 29, 50]),
these fluctuations are likely to have functional consequences. Indeed,
comparable fluctuations in connection strengths are observed when
measured directly [51, 52, 53]. Furthermore, fluctuations tend to occur simultaneously in pre- and postsynaptic scaffolds of the same synapses [43, further supporting their functional significance.
As
synapses in the aforementioned studies were embedded in active
networks, their fluctuations reflect both spontaneous and
activity-history-driven processes. Yet even when spontaneous activity is
suppressed or eliminated altogether, fluctuations persist [3, 6, 17, 40, 54],
although their characteristics can change (see below). These results
suggest that in mature networks, as in developing networks, spontaneous
synaptic remodeling is significant.
Relative Contributions of Specific Activity Histories and Spontaneous Processes to Synaptic Remodeling
The
presence of significant activity-independent synaptic remodeling raises
an important question: what are the relative contributions of specific
activity histories and spontaneous processes to synaptic remodeling?
This question might be addressed by comparing the magnitudes of changes
they induce. As a rough estimate, changes to PSD sizes following
experimental paradigms that induce long-term potentiation (measured
using fluorescently tagged variants of PSD-95 and Homer1b) (e.g., [30, 31]) seem to be of magnitude similar to that of baseline PSD size fluctuations observed in vivo using similar approaches [47, 49].
However, these comparisons are made among different experimental
systems and conditions (e.g., rat hippocampus CA1 pyramidal cells in
cultured slices [30, 31] and mouse layer 2/3 cortical pyramidal cells in vivo [47, 49]).
The fluctuations observed in the latter studies represent the sum of both activity-dependent and -independent processes. As a step toward separating their respective contributions, the effects of suppressing activity were examined. Such experiments (in culture) [17, 54] have shown that suppressing all spontaneous activity reduces the magnitude of glutamatergic synapse size fluctuations, but only by a factor of about two. Interestingly, little to no change was observed for GABAergic synapses [6, 40].
A more direct estimate of the relative contributions is obtained by comparing pairs of synapses formed between the same axons and dendrites (see Figure 2 for an illustration of this approach). Such pairs share common activation histories that should drive similar remodeling in the two synapses. Conversely, size or remodeling differences between synapses belonging to the same pair would reflect spontaneous, activity-history-independent processes occurring autonomously at each synapse. This rationale was applied in several recent studies using electron microscopy reconstructions of hippocampal tissue [55, a fully reconstructed volume of mouse neocortex [56, and long-term imaging of cortical neurons in culture [54. In general, it was found that such synapse pairs tend to be more similar than randomly chosen pairs. However, remodeling covariance over 48-h periods, as well as correlation coefficients of instantaneous sizes (spine volumes and PSD areas) were relatively small (∼0.25–0.35). This led to an estimate that at most 40% of synaptic remodeling could be attributed to specific activity histories [54.
Collectively
these findings suggest that the contributions of spontaneous processes
and specific activity histories to synaptic remodeling are of similar
magnitudes.
The fluctuations observed in the latter studies represent the sum of both activity-dependent and -independent processes. As a step toward separating their respective contributions, the effects of suppressing activity were examined. Such experiments (in culture) [17, 54] have shown that suppressing all spontaneous activity reduces the magnitude of glutamatergic synapse size fluctuations, but only by a factor of about two. Interestingly, little to no change was observed for GABAergic synapses [6, 40].
A more direct estimate of the relative contributions is obtained by comparing pairs of synapses formed between the same axons and dendrites (see Figure 2 for an illustration of this approach). Such pairs share common activation histories that should drive similar remodeling in the two synapses. Conversely, size or remodeling differences between synapses belonging to the same pair would reflect spontaneous, activity-history-independent processes occurring autonomously at each synapse. This rationale was applied in several recent studies using electron microscopy reconstructions of hippocampal tissue [55, a fully reconstructed volume of mouse neocortex [56, and long-term imaging of cortical neurons in culture [54. In general, it was found that such synapse pairs tend to be more similar than randomly chosen pairs. However, remodeling covariance over 48-h periods, as well as correlation coefficients of instantaneous sizes (spine volumes and PSD areas) were relatively small (∼0.25–0.35). This led to an estimate that at most 40% of synaptic remodeling could be attributed to specific activity histories [54.
Characteristics, Sources, and Implications of Spontaneous Remodeling Processes
It
appears that spontaneous remodeling is a central, inherent feature of
CNS synapses, which merits in-depth analyses of its characteristics,
sources, implications, and physiological importance. We mention a few of
these here.
In the framework of synaptic plasticity, significance is attached to the specific configuration of synaptic strengths (synaptic weights) in a network. Therefore, it is important to characterize the rates at which spontaneous remodeling might ‘erode’ configurations of synaptic input strengths (Figure 3). Imaging studies in primary culture indicate that synaptic configurations erode significantly over timescales of a few days [3, 4, 5, 6, 27, 38]. This becomes evident when synaptic size is plotted against initial size at increasingly greater time intervals, as illustrated in Figure 3B [6, 27]. Not only are correlations reduced, but linear regression fits gradually become shallower while their offset terms grow larger. Suppressing network activity slows this process, but only by a factor of about two [3. Interestingly, erosion rates of GABAergic synapse configurations seem to be insensitive to activity levels and slower than those of glutamatergic synapses in the same neurons [6.
Given
enough time, will synaptic configurations erode completely? In studies
published to date, monotonic erosion of synaptic configurations was
reported [3, 4, 5, 6, 27, 38];
however, over the timescales of such experiments, configurations did
not erode entirely. Perhaps, given sufficient time, they would (Figure 3D,
red broken lines). Alternatively however, they might conserve a
semblance of their original configuration even after long times (Figure 3D, green broken lines). This could reflect limited mixing due to sizes fluctuating around synapse-specific set points (Figure 3E) defined, for example, by relatively stable core scaffolds [4, 16].
Another possibility is that synaptic tenacity varies in the population
such that the sizes of some synapses fluctuate much more than others. At
present the extent of mixing and the heterogeneity of synaptic tenacity
remain unknown. The rates at which synaptic configurations erode in vivo and how these might change with age [39 (see also [57)
or in association with various neurological disorders are not known
either. These are crucial questions that call for very long
measurements, particularly in vivo (see Outstanding Questions).
Finally, the degree to which the observed erosion of synaptic
configurations ultimately affects network function is unclear, but
potential links between such erosion and memory decay [17 or declining mental abilities are plausible.
Although synaptic configurations deteriorate, the same analyses reveal other properties that emerge as stable invariants at the population level.
First, while single-synapse properties (e.g., spine volumes, PSD molecule contents) fluctuate, distributions of the same properties can be very stable [3, 6, 27, 38]. Such distributions quantify the proportions of synapses of different sizes in the population; distribution stability implies that fluctuations in synaptic sizes are not merely diffusive (which would lead to the gradual broadening of size distributions) but reflect a dynamic equilibrium in which synaptic sizes are somehow constrained (Box 1). One way to appreciate the presence of such constraints is by plotting changes in synaptic sizes against their initial sizes at later times (Figure 3C). Such plots highlight the tendency of small synapses to grow larger and of large synapses to become smaller [3, 6, 27], with the magnitude such changes increasing with time.
Second,
the shapes of distributions within synaptic populations (e.g.,
distributions of release probability, synaptic currents, spine volumes,
PSD molecule and receptor contents) tend to be skewed; that is, to have a
tail of particularly large or strong synapses (e.g., [6, 18, 19, 27, 58, 59]; reviewed in [60). It has been suggested that such distributions might be a consequence of synaptic plasticity in several forms [61. Skewed distributions emerge, however, even when activity is suppressed throughout network development [18, 19],
highlighting once again the significant contributions of spontaneous
remodeling and its ability to generate full repertoires of synapses at
correct proportions.
The realization that size fluctuations play crucial roles in synaptic remodeling has motivated several groups to explore this topic through computational modeling [17, 27, 29, 38]. These models faithfully capture many of the experimental results and provide some insights into their potential origin. For example, they offer an explanation of how synaptic size fluctuations can give rise to skewed, stable size distributions (Box 1). They also provide an explanation for experimental findings showing that scaling of synaptic size distributions can occur without overt, uniform multiplicative scaling of individual synapse sizes [3, 27] (Box 2).
A more recent computational study [62
has gone one step further, showing that all of these properties – size
fluctuations and stable and skewed size distributions, as well as their
scaling – can emerge naturally from stochastic assimilation and removal
of synaptic molecules at synaptic sites, provided the two processes
exhibit cooperativity. This model also recapitulates the internal
spatial organization of synapses in the form of dynamic nanodomains
(e.g., [63, 64]).
A more detailed model, where removal and aggregation are implemented by
lateral diffusion of receptors bound to scaffold proteins, also
provides good fits to measured distributions [65.
Since distribution properties can be described equally well by
alternative mesoscopic models, further experimental and theoretical work
is required to identify the key biophysical processes involved.
In the framework of synaptic plasticity, significance is attached to the specific configuration of synaptic strengths (synaptic weights) in a network. Therefore, it is important to characterize the rates at which spontaneous remodeling might ‘erode’ configurations of synaptic input strengths (Figure 3). Imaging studies in primary culture indicate that synaptic configurations erode significantly over timescales of a few days [3, 4, 5, 6, 27, 38]. This becomes evident when synaptic size is plotted against initial size at increasingly greater time intervals, as illustrated in Figure 3B [6, 27]. Not only are correlations reduced, but linear regression fits gradually become shallower while their offset terms grow larger. Suppressing network activity slows this process, but only by a factor of about two [3. Interestingly, erosion rates of GABAergic synapse configurations seem to be insensitive to activity levels and slower than those of glutamatergic synapses in the same neurons [6.
Although synaptic configurations deteriorate, the same analyses reveal other properties that emerge as stable invariants at the population level.
First, while single-synapse properties (e.g., spine volumes, PSD molecule contents) fluctuate, distributions of the same properties can be very stable [3, 6, 27, 38]. Such distributions quantify the proportions of synapses of different sizes in the population; distribution stability implies that fluctuations in synaptic sizes are not merely diffusive (which would lead to the gradual broadening of size distributions) but reflect a dynamic equilibrium in which synaptic sizes are somehow constrained (Box 1). One way to appreciate the presence of such constraints is by plotting changes in synaptic sizes against their initial sizes at later times (Figure 3C). Such plots highlight the tendency of small synapses to grow larger and of large synapses to become smaller [3, 6, 27], with the magnitude such changes increasing with time.
Models of Synaptic Size Dynamics
The realization that size fluctuations play crucial roles in synaptic remodeling has motivated several groups to explore this topic through computational modeling [17, 27, 29, 38]. These models faithfully capture many of the experimental results and provide some insights into their potential origin. For example, they offer an explanation of how synaptic size fluctuations can give rise to skewed, stable size distributions (Box 1). They also provide an explanation for experimental findings showing that scaling of synaptic size distributions can occur without overt, uniform multiplicative scaling of individual synapse sizes [3, 27] (Box 2).
Synaptic Scaling as a Population-Level Phenomenon
Concluding Remarks
The findings summarized above indicate that synaptic tenacity is inherently limited [3, 4, 6, 43, 54] or, using the terminology of Rumpel, Loewenstein, and others [34, 35],
that synapses are intrinsically ‘volatile’. How can the notions of
synaptic plasticity and synaptic volatility be reconciled? How can
properties of individual neurons, such as sensory tuning and motor and
place representation, as well as higher-level functions remain invariant
if directed and spontaneous synaptic changes are, as suggested by
various studies, of similar magnitude? Potential answers to these
questions have been discussed in several recent reviews [33, 34, 35] and a subset of possible resolutions is summarized briefly in Box 3.
Irrespective of interpretation, however, it is clear that the ideal
notion of perfectly tenacious synapses is not supported by experimental
findings (nor is it particularly plausible from a biological standpoint,
given the challenges to synaptic tenacity discussed above).
Furthermore, spontaneous synaptic remodeling has important consequences
that are not fully understood and are not always intuitive (see
Outstanding Questions). When it comes to cognitive functions, long-term
memory is one area where the notion of synaptic volatility raises
perhaps some of the most challenging questions. In light of findings
discussed in this Opinion article, and possibly others [34, 66, 67],
age-old notions concerning relationships between histories of
‘elementary brain-processes’, connection strengths, and memory traces
might need to be revisited; put differently (to paraphrase James [1), modern science might need to improve on this explanation.
Memory in the Presence of Limited Synaptic Tenacity
+
Outstanding Questions
Published online: December 21, 2017
© 2017 The Authors. Published by Elsevier Ltd.
Access this article on
No comments:
Post a Comment